The development, study, and experimental validation of cluster and grid middlewares are combersome, time and resource consuming. Researchers need to deploy their middlewares on multiple nodes to test and evaluate their modifications. But these middlewares are destine to be used in production with hundreds or thousands of machines, and deploying at this scale to perform tests and evaluations is too costly. In this paper, we investigate folding techniques to represent a full scale cluster on less physical machines. We evaluate the performance of a distributed I/O benchmark application with respect to the level of folding to determine guidelines for future folded experiments containing a distributed file-system.